[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] Re: [PATCH 2/2] Add physical CPU hotplug support in PV_ops dom0



On 09/24/09 17:27, Jiang, Yunhong wrote:
> Do you mean logical offlining or physical offlining? I think logical 
> offlining is tested long before. For physical offlining, I implement it but I 
> have no platform to test still.
>
> In fact , it is  a tricky to physical offlining a cpu (or, more precisely 
> speaking, socket) because that means we need offline all memory behind a 
> socket. But it can be used for socket migration, i.e. put down the old CPU 
> and bring up a spare CPU, in that situation, we need CPU offlining. We are 
> still working on socket migration.
>   

I see.  I guess it makes it hard when the CPUs are also the memory
controllers.

> Yes, also MCE.
> And, are you sure currently the microcode driver really not called in hotplug 
> of vcpus? Will vcpu online not trigger the CPU_online notifier?
>   

I would expect vcpu hotplug will call the usual notifiers as expected,
but drivers which really care about pcpus don't have much use for those
notifications.
>> This could do with some clarification.  Is this case that a newly
>> added pcpu already appears to be online?
>>     
> Because the vIRQ notification for pcpu hotplug from xen hypervisor is async, 
> so maybe it happens after the pcpu is onlined by user. (i.e., the online 
> notification and hotplug notification is merged)
>   

I'm not sure I follow.  Are you saying this is an expected race, and
that this printk is just for debugging?

>> __init?
>>
>> Also I prefer names of the form subsystem_action, so
>> xen_pcpu_info_init().
>>     
> Sorry, what do you mean of subsystem_action? I didn't find definition for it.
>   

I meant "<subsystem>_<action>()".

>> So when does this interrupt get raised?  Is the full flow:
>>
>>   1. ACPI raises SCI
>>   2. dom0 catches that and gets the new pcpu event
>>   3. dom0 notifies Xen that the pcpu exists
>>   4. xen interrupts dom0 saying that a new pcpu exists
>>   5. dom0 processes the list and adds it to sysfs
>>   6. usermode onlines the cpu for Xen's use
>>     
> And there is also step 7, Xen interrupt dom0 saying a new pCPU is onlined. 
> Through step 7, dom0 will have the life cycle of the pCPU.
>   

OK, I see.

    J

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.